Pith. sign in

REVIEW 4 major objections 5 minor 1 cited by

Exp-Graph: How Connections Learn Facial Attributes in Graph-based Expression Recognition

T0 review · 4 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

Pith's one-line read Exp-Graph represents a face as a graph of landmarks, with edges set by spatial proximity and vision-transformer feature similarity, and reports recognition accuracies of 98.09%, 79.01%, and 56.39% on three benchmark datasets.

desk verdict A reasonable graph-based FER idea, but the experimental tables contradict each other so badly that the headline accuracies are not credible. read the letter →

arxiv 2507.14608 v1 pith:W5DLQB3D submitted 2025-07-19 cs.CV cs.AIcs.LG

classification cs.CVcs.AIcs.LG
keywords facialexpressionrecognitiongraphconvolutionalnetworksvisiontransformerlandmarksdynamicconstructionaffectivecomputingattribute
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

Exp-Graph's goal is to show that facial expression recognition improves when facial attributes are represented as a graph rather than a flat feature vector or a fixed template. The graph's vertices are facial landmarks; an edge survives only when two landmarks are spatially close and their local appearance patches, encoded by a pre-trained vision transformer, are similar. A graph convolutional network then propagates features along these learned connections and classifies the expression. If the reported numbers hold, the approach would transfer from controlled laboratory data to in-the-wild video, which matters for affective computing applications such as face animation, surveillance, and medical analysis.

What carries the argument

The central object is the dynamic facial-attribute graph, with landmarks as vertices and edges learned from both geometry and appearance. The raw edge weight is $A_{ij} = K(x_i,x_j)/e^{\|p_i-p_j\|}$, where $x_i$ is the ViT feature of the patch around landmark $p_i$; the threshold $T_s = \mu_K + \tau\sigma_K$ then keeps only connections above it. The GCN consumes the binary adjacency through the symmetrically normalized form $\hat{A} = \tilde{D}^{-1/2}(A+I)\tilde{D}^{-1/2}$ and updates node features by $H^{(l+1)} = \sigma(\hat{A}H^{(l)}W^{(l)})$. That mechanism lets local appearance and landmark geometry jointly decide which facial attributes are related before the classifier votes on the expression.

What would settle it

Two checks would settle the claim: rerun the described pipeline on Oulu-CASIA with a fixed subject-independent split, patch size 70×70, threshold $\tau=0.5$, and no data augmentation, and see whether accuracy reproduces 98.09%; then compare the binary adjacency matrices for disgust and sadness images with nearly identical landmark geometry, because if the edge sets are identical, the appearance-similarity term is not doing the distinguishing work the paper assigns it.

Watch

Extended reading notes

Core claim

Exp-Graph's central claim is that the connections between facial landmarks carry much of the information needed to tell expressions apart, not just landmark locations or raw appearance. The paper builds one graph per face: the adjacency matrix $A_{ij} = K(x_i,x_j)/e^{\|p_i-p_j\|}$ combines ViT feature similarity $K(x_i,x_j)$ with landmark proximity through an exponential distance term, and a threshold $T_s = \mu_K + \tau\sigma_K$ binarizes the matrix. A GCN with self-loops then propagates and projects node features, and cross-entropy loss trains the classifier. On this construction the paper reports 98.09% accuracy on Oulu-CASIA, 79.01% on eNTERFACE05, and 56.39% on AFEW, surpassing the compared baselines on each dataset.

Load-bearing premise

The load-bearing premise is that the reported accuracies came from the described method under standard, fixed evaluation settings; the paper does not specify the train/test split, and identical numeric rows appear repeated across different datasets in the internal tables, so the numbers cannot be reproduced from the text alone.

Editorial extensions

If this is right

  • If the reported accuracies hold, a dynamic graph built from landmark proximity and ViT patch similarity generalizes across laboratory and in-the-wild expression datasets without a fixed, predefined graph structure.
  • The threshold $\tau$ and patch size are explicit controls on graph sparsity, and the paper reports that the best values are dataset-dependent: patch 70×70 works best on Oulu-CASIA while patch 30×30 works best on eNTERFACE05.
  • Under the paper's evaluation, the ViT+GCN combination outperforms the same feature encoder paired with graph attention, and it exceeds the compared baselines on all three datasets.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The same edge-construction rule could transfer to other landmark-based tasks, such as action-unit detection or micro-expression spotting, wherever pairwise relationships among facial keypoints carry semantic weight.
  • A learned or adaptive threshold could replace the manually chosen $\tau$ and patch size, eliminating the dataset-specific hand-selection the paper currently reports.
  • If the AFEW result holds up, the landmark-graph representation might serve as a lightweight, privacy-preserving alternative to full-image classifiers, since it consumes geometry and local patches rather than raw pixels.
  • The paper does not test sensitivity to landmark detection errors; adding controlled noise to landmark coordinates would show how much of the reported accuracy depends on perfect keypoints.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 5 minor

Summary. The paper proposes Exp-Graph, a facial expression recognition framework that builds a graph whose nodes are Dlib facial landmarks and whose edges are derived from the proximity of landmarks and the similarity of local ViT patch features via Eqs. (1)-(2), then classifies expressions with a graph convolutional network. The authors report state-of-the-art accuracies of 98.09% on Oulu-CASIA, 79.01% on eNTERFACE05, and 56.39% on AFEW, and present threshold sweeps, patch-size sweeps, backbone comparisons, and t-SNE visualizations.

Significance. The core idea, dynamically connecting facial landmarks using both geometry and local appearance encoded by a vision transformer, is plausible and could be a useful contribution to graph-based facial expression recognition. The paper contains algorithmic descriptions (Algorithms 1 and 2), uses public benchmarks, and reports multiple ablations, which are positive elements. However, the experimental evidence is internally inconsistent and under-specified: identical numeric rows appear for different datasets, the threshold definition in Eq. (2) is incompatible with the threshold values used in the tables, and no single experimental configuration supports the headline accuracies. As a result, the claimed results cannot be reproduced or verified from the manuscript, and the significance of the method cannot be assessed as it stands.

major comments (4)
  1. [§IV, Table V] Table V reports identical Loss, F1, WAR, and UAR values for Oulu-CASIA and eNTERFACE05 at every listed threshold, for example 1.34/68.11/69.88/85.51 at Th=0.30 and 1.21/84.00/84.00/92.00 at Th=0.50. These are different datasets with different numbers of subjects, different modalities, and different class distributions, so identical performance metrics are implausible. This directly undermines the threshold-sweep analysis on which the paper's performance claims rely.
  2. [§IV, Tables VIII and IX] The AFEW rows in Table IX at Th=0.30 (62.67/59.27/62.67/82.50), Th=0.70 (25.71/9.03/25.71/82.73), and Th=0.90 (21.08/11.06/21.08/74.74) exactly duplicate the eNTERFACE05 rows in Table VIII. Meanwhile the AFEW Th=0.50 row (56.39) is the value quoted in the abstract and Table IV, while the text states that AFEW's optimum is at Th=0.30 with accuracy 62.67. These duplicated and mutually contradictory numbers make the AFEW results unusable as published.
  3. [§III-A and §IV, Eq. (2) and Tables V–IX] The threshold is defined in Eq. (2) as Ts = mu_K + tau*sigma_K, which makes tau a distribution-dependent multiplier, but the experiments treat Th = tau as an absolute threshold in [0.20, 0.90] across Tables V–IX and Fig. 6, with no actual Ts values reported. These two interpretations are incompatible, so the threshold values in the tables are ambiguous and the graph-construction procedure cannot be reproduced from the paper. In addition, the headline accuracies are not tied to a single configuration: Oulu-CASIA's 98.09% appears only in Table X at patch size 70x70 with no threshold stated, while Table VII reports a best Oulu-CASIA accuracy of 91.09% at Th=0.50 without augmentation; eNTERFACE05's 79.01% in Table III differs from the 79.06% reported as optimal in Table XI; and AFEW's 56.39% in Table IV conflicts with the 62.67% reported for AFEW in Table IX.
  4. [§IV, experimental setup and Tables II–IV] The manuscript never states the train/validation/test split for any dataset, nor whether classification is performed on frames or on video-level predictions, despite Table I describing eNTERFACE05 as visual-audio and AFEW as in-the-wild video. It also does not specify whether data augmentation is used for the headline numbers; Table VII is explicitly labeled as without augmentation, while no augmented run is reported. Without this information, the accuracies in Tables II–IV cannot be reproduced, and the comparison with state-of-the-art methods is not placed on a common protocol.
minor comments (5)
  1. [§III-A, Eq. (1)-(2)] The symbol A is used both for the unthresholded similarity-distance matrix in Eq. (1) and for the binarized adjacency matrix in Eq. (2); using different symbols would avoid confusion.
  2. [§IV, Table III title] The table title uses ENTERFACE05, while the dataset name elsewhere is eNTERFACE05; the spelling should be consistent.
  3. [§IV, reference [23]] The text states that EfficientNetB0 is used as a base model and cites [23], but [23] is the Graph Attention Networks paper; EfficientNetB0 should be cited to Tan and Le (ref. [86]).
  4. [§IV, Fig. 10 and Fig. 11] The sentence beginning 'Additionally, 10b presents t-SNE visualizations...' appears to refer to Fig. 11b rather than Fig. 10b, since Fig. 10 shows bar charts.
  5. [§I, Introduction] The sentence 'However, facial expression recognition using GCNs facial challenges...' appears to be missing a verb, likely 'faces'; please revise.

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity: Exp-Graph's accuracy claims are empirical, and the only self-citations are background references that do not carry the argument.

full rationale

This paper is an empirical systems paper: it proposes a graph-based FER pipeline and reports benchmark accuracies. There is no claimed derivation chain in which a prediction follows from a fitted input by construction. The adjacency construction in Eqs. (1)-(2) combines ViT feature similarity with landmark proximity and a threshold; this is a stated design choice, not a result derived from the reported accuracies. The headline numbers come from model evaluation, not from a parameter that was fitted to the same statistic being predicted. Threshold and patch-size selection are hyperparameter choices reported per dataset, which is standard practice and does not constitute 'fitted input called prediction' unless the paper claimed to predict them. The self-citations [17]-[19] appear only in the introduction's background remark that appearance-only models cannot exploit structure when training data is limited; they are not load-bearing for the central claim and no uniqueness theorem or forced-choice argument is imported from them. The serious internal inconsistencies in the experimental tables (identical Oulu and eNTERFACE05 rows in Table V, duplicated eNTERFACE05 rows in Table IX, and the unreported configuration behind the 98.09% figure) are reproducibility and soundness concerns, not circularity: they do not show that the output is equivalent to the input by definition. Accordingly, the circularity score is low, reflecting only the presence of minor non-load-bearing self-citations.

Assumptions & free parameters 2 free parameters · 4 assumptions · 0 invented entities

The central claim rests on a small number of domain assumptions about landmark reliability and the informativeness of the hand-designed edge kernel. The two major free parameters, threshold and patch size, are tuned per dataset and not robustly justified. No new entities are introduced.

free parameters (2)
  • Threshold tau = 0.5 for Oulu, 0.3 for eNTERFACE/AFEW per tables, but inconsistent
    The threshold controls graph sparsity and is tuned per dataset. The paper's formula in Eq. (2) is inconsistent with the experimental 'Th' values, making the exact parameter choice ambiguous.
  • Patch size = 70x70 for Oulu, 30x30 for eNTERFACE, AFEW not specified
    Patch size determines the local context extracted around each landmark and is swept per dataset. The chosen patch size strongly affects the reported accuracy, as shown in Tables X and XI.
assumptions (4)
  • domain assumption Facial expression is encoded in the geometric structure and local appearance around facial landmarks.
    This motivates building graphs over landmarks and using ViT patch features. Stated in Section I and III-A.
  • domain assumption Dlib landmark detection reliably localizes facial attributes in all test images, including in-the-wild AFEW.
    The graph construction depends entirely on landmark coordinates, yet no failure analysis or handling of missed faces is provided.
  • domain assumption Cosine similarity of ViT features weighted by inverse exponential distance, then thresholded, yields more informative edges than fixed anatomical adjacency.
    This is the core modeling assumption, but the paper does not compare alternative edge functions or learn edge weights, despite claiming to learn connections.
  • standard math Standard GCN propagation with symmetric normalization is valid for this graph.
    Equations (3) and (4) follow Kipf and Welling's GCN formulation, which is accepted in the literature.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Exp-Graph: How Connections Learn Facial Attributes in Graph-based Expression Recognition." pith.science (2026). https://pith.science/paper/W5DLQB3D

@misc{pith2026250714608,
  author       = {Pith},
  title        = {Pith review of: Exp-Graph: How Connections Learn Facial Attributes in Graph-based Expression Recognition},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/W5DLQB3D}},
  note         = {Machine review of arXiv:2507.14608}
}
read the original abstract

Facial expression recognition is crucial for human-computer interaction applications such as face animation, video surveillance, affective computing, medical analysis, etc. Since the structure of facial attributes varies with facial expressions, incorporating structural information into facial attributes is essential for facial expression recognition. In this paper, we propose Exp-Graph, a novel framework designed to represent the structural relationships among facial attributes using graph-based modeling for facial expression recognition. For facial attributes graph representation, facial landmarks are used as the graph's vertices. At the same time, the edges are determined based on the proximity of the facial landmark and the similarity of the local appearance of the facial attributes encoded using the vision transformer. Additionally, graph convolutional networks are utilized to capture and integrate these structural dependencies into the encoding of facial attributes, thereby enhancing the accuracy of expression recognition. Thus, Exp-Graph learns from the facial attribute graphs highly expressive semantic representations. On the other hand, the vision transformer and graph convolutional blocks help the framework exploit the local and global dependencies among the facial attributes that are essential for the recognition of facial expressions. We conducted comprehensive evaluations of the proposed Exp-Graph model on three benchmark datasets: Oulu-CASIA, eNTERFACE05, and AFEW. The model achieved recognition accuracies of 98.09\%, 79.01\%, and 56.39\%, respectively. These results indicate that Exp-Graph maintains strong generalization capabilities across both controlled laboratory settings and real-world, unconstrained environments, underscoring its effectiveness for practical facial expression recognition applications.

Figures

Figures reproduced from arXiv: 2507.14608 by the authors.

Figure 1
Figure 1. System architecture of facial expression recognition using Exp-Graph [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 3
Figure 3. Geometry alone is insufficient. [Best shown in color] [PITH_FULL_IMAGE:figures/full_fig_p002_3.png] view at source ↗
Figure 4
Figure 4. An illustration of feature extraction of facial units using vision [PITH_FULL_IMAGE:figures/full_fig_p004_4.png] view at source ↗
Figures from the paper (5 more)
Figure 5
Figure 5. Figure 5: An outline of the proposed framework for detecting and recognizing facial expressions. The exp-Graph framework is composed of two primary steps: [PITH_FULL_IMAGE:figures/full_fig_p005_5.png]
Figure 6
Figure 6. Figure 6: Test results across thresholds (Th = τ) for three datasets. [Best shown in color] (92%). Performance degrades significantly at higher threshold [PITH_FULL_IMAGE:figures/full_fig_p007_6.png]
Figure 7
Figure 7. Figure 7: Test result on the Oulu-CASIA dataset across models. [Best shown [PITH_FULL_IMAGE:figures/full_fig_p008_7.png]
Figure 9
Figure 9. Figure 9: Visualization of learned graphs for the Oulu-CASIA dataset sample [PITH_FULL_IMAGE:figures/full_fig_p009_9.png]
Figure 10
Figure 10. Figure 10: Test results across patch sizes for two FER datasets at a threshold [PITH_FULL_IMAGE:figures/full_fig_p010_10.png]

Discussion (0). Sign in to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Cranio-ID: Graph-Based Craniofacial Identification via Automatic Landmark Annotation in 2D Multi-View X-rays

    cs.CV 2025-11 reject novelty 4.0 of 10

    Cranio-ID retrieves faces from skull X-rays and sketches by combining YOLO-pose landmark detection with graph-based cross-attention and optimal-transport matching.

Reference graph

Works this paper leans on

86 extracted references · 74 canonical work pages · cited by 1 Pith paper

  1. [1]

    Gabor-based kernel pca with doubly nonlinear mapping for face recognition with a single face image,

    X. Xie and K.-M. Lam, “Gabor-based kernel pca with doubly nonlinear mapping for face recognition with a single face image,”IEEE Transac- tions on Image Processing, vol. 15, no. 9, pp. 2481–2492, 2006

  2. [2]

    Dynamic texture recognition using local binary patterns with an application to facial expressions,

    G. Zhao and M. Pietikainen, “Dynamic texture recognition using local binary patterns with an application to facial expressions,”IEEE Trans- actions on Pattern Analysis and Machine Intelligence, vol. 29, no. 6, pp. 915–928, 2007

  3. [3]

    Facial expression recognition based on local binary patterns: A comprehensive study,

    C. Shan, S. Gong, and P. W. McOwan, “Facial expression recognition based on local binary patterns: A comprehensive study,”Image and vision Computing, vol. 27, no. 6, pp. 803–816, 2009

  4. [4]

    Emotion recognition using phog and lpq features,

    A. Dhall, A. Asthana, R. Goecke, and T. Gedeon, “Emotion recognition using phog and lpq features,” in2011 IEEE International Conference on Automatic Face & Gesture Recognition (FG). IEEE, 2011, pp. 878–883

  5. [5]

    Facial expression recognition and analysis: a comparison study of feature descriptors,

    C. F. Liew and T. Yairi, “Facial expression recognition and analysis: a comparison study of feature descriptors,”IPSJ transactions on computer vision and applications, vol. 7, pp. 104–120, 2015

  6. [6]

    Facial expression recognition in video with multiple feature fusion,

    J. Chen, Z. Chen, Z. Chi, and H. Fu, “Facial expression recognition in video with multiple feature fusion,”IEEE Transactions on Affective Computing, vol. 9, no. 1, pp. 38–50, 2016

  7. [7]

    Facial expression recognition using iterative fusion of mo-hog and deep features,

    H. Wang, S. Wei, and B. Fang, “Facial expression recognition using iterative fusion of mo-hog and deep features,”The Journal of Super- computing, vol. 76, no. 5, pp. 3211–3221, 2020

  8. [8]

    Context-aware emotion recognition networks,

    J. Lee, S. Kim, S. Kim, J. Park, and K. Sohn, “Context-aware emotion recognition networks,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2019, pp. 10 143– 10 152

Show all 86 references
  1. [9]

    Spatial- temporal graphs plus transformers for geometry-guided facial expression recognition,

    R. Zhao, T. Liu, Z. Huang, D. P. Lun, and K.-M. Lam, “Spatial- temporal graphs plus transformers for geometry-guided facial expression recognition,”IEEE Transactions on Affective Computing, vol. 14, no. 4, pp. 2751–2767, 2022

  2. [10]

    Emoticon: Context-aware multimodal emotion recogni- tion using frege’s principle,

    T. Mittal, P. Guhan, U. Bhattacharya, R. Chandra, A. Bera, and D. Manocha, “Emoticon: Context-aware multimodal emotion recogni- tion using frege’s principle,” inProceedings of the IEEE/CVF Confer- ence on Computer Vision and Pattern Recognition (CVPR), 2020, pp. 14 234–14 243

  3. [11]

    Deep-emotion: Facial ex- pression recognition using attentional convolutional network,

    S. Minaee, M. Minaei, and A. Abdolrashidi, “Deep-emotion: Facial ex- pression recognition using attentional convolutional network,”Sensors, vol. 21, no. 9, p. 3046, 2021

  4. [12]

    A discriminative feature learning approach for deep face recognition,

    Y . Wen, K. Zhang, Z. Li, and Y . Qiao, “A discriminative feature learning approach for deep face recognition,” inComputer vision–ECCV 2016: 14th European conference, amsterdam, the netherlands, October 11–14, proceedings, part VII 14. Springer, 2016, pp. 499–515

  5. [13]

    Reliable crowdsourcing and deep locality- preserving learning for expression recognition in the wild,

    S. Li, W. Deng, and J. Du, “Reliable crowdsourcing and deep locality- preserving learning for expression recognition in the wild,” inProceed- ings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2017, pp. 2852–2861

  6. [14]

    Island loss for learning discriminative features in facial expression recognition,

    J. Cai, Z. Meng, A. S. Khan, Z. Li, J. O’Reilly, and Y . Tong, “Island loss for learning discriminative features in facial expression recognition,” in 2018 13th IEEE International Conference on Automatic Face & Gesture Recognition (FG 2018). IEEE, 2018, pp. 302–309

  7. [15]

    Facial expression recog- nition in the wild using multi-level features and attention mechanisms,

    Y . Li, G. Lu, J. Li, Z. Zhang, and D. Zhang, “Facial expression recog- nition in the wild using multi-level features and attention mechanisms,” IEEE Transactions on Affective Computing, vol. 14, no. 1, pp. 451–462, 2020

  8. [16]

    Deep multi- task learning for facial expression recognition and synthesis based on selective feature sharing,

    R. Zhao, T. Liu, J. Xiao, D. P. Lun, and K.-M. Lam, “Deep multi- task learning for facial expression recognition and synthesis based on selective feature sharing,” in2020 25th Proceedings of the International Conference on Pattern Recognition (ICPR). IEEE, 2021, pp. 4412– 4419

  9. [17]

    Graph formulation of video activities for abnormal activity recognition,

    D. Singh and C. K. Mohan, “Graph formulation of video activities for abnormal activity recognition,”Pattern Recognition, vol. 65, pp. 265– 272, 2017

  10. [18]

    Graphlime: Local interpretable model explanations for graph neural networks,

    Q. Huang, M. Yamada, Y . Tian, D. Singh, and Y . Chang, “Graphlime: Local interpretable model explanations for graph neural networks,”IEEE Transactions on Knowledge and Data Engineering, vol. 35, no. 7, pp. 6968–6972, 2022

  11. [19]

    Graph representation for weakly-supervised spatio-temporal action detection,

    D. Singh, “Graph representation for weakly-supervised spatio-temporal action detection,” inProceedings of the International Joint Conference on Neural Networks (IJCNN). IEEE, 2023, pp. 1–9

  12. [20]

    An image is worth 16x16 words: Transformers for image recognition at scale,

    A. Dosovitskiy, L. Beyer, A. Kolesnikov, D. Weissenborn, X. Zhai, T. Unterthiner, M. Dehghani, M. Minderer, G. Heigold, S. Gellyet al., “An image is worth 16x16 words: Transformers for image recognition at scale,”arXiv preprint arXiv:2010.11929, 2020

  13. [21]

    Facial expression recognition from occluded images using deep convolution neural network with vision transformer,

    M. Li, S. Tu, and S. u. Rehman, “Facial expression recognition from occluded images using deep convolution neural network with vision transformer,” inInternational Conference on Image and Graphics. Springer, 2023, pp. 289–299. JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. X, AUGU...

  14. [22]

    Facial expression recognition on the high aggregation subgraphs,

    T. Liu, J. Li, J. Wu, B. Du, J. Chang, and Y . Liu, “Facial expression recognition on the high aggregation subgraphs,”IEEE Transactions on Image Processing, 2023

  15. [23]

    Graph attention networks,

    P. Veli ˇckovi´c, G. Cucurull, A. Casanova, A. Romero, P. Lio, and Y . Ben- gio, “Graph attention networks,”arXiv preprint arXiv:1710.10903, 2017

  16. [24]

    Visual transformers: Token- based image representation and processing for computer vision,

    B. Wu, C. Xu, X. Dai, A. Wan, P. Zhang, Z. Yan, M. Tomizuka, J. Gonzalez, K. Keutzer, and P. Vajda, “Visual transformers: Token- based image representation and processing for computer vision,” 2020

  17. [25]

    Facial expression recognition with visual transformers and attentional selective fusion,

    F. Ma, B. Sun, and S. Li, “Facial expression recognition with visual transformers and attentional selective fusion,”IEEE Transactions on Affective Computing, vol. 14, no. 2, pp. 1236–1248, 2021

  18. [26]

    Geometry guided pose- invariant facial expression recognition,

    F. Zhang, T. Zhang, Q. Mao, and C. Xu, “Geometry guided pose- invariant facial expression recognition,”IEEE Transactions on Image Processing, vol. 29, pp. 4445–4460, 2020

  19. [27]

    Hypergraph- guided disentangled spectrum transformer networks for near-infrared facial expression recognition,

    B. Luo, H. Wang, J. Wang, J. Zhu, X. Zhao, and Y . Gao, “Hypergraph- guided disentangled spectrum transformer networks for near-infrared facial expression recognition,” inProceedings of the AAAI Conference on Artificial Intelligence, vol. 38, no. 9, 2024, pp. 10 101–10 109

  20. [28]

    Semi-supervised classification with graph convolutional networks,

    T. N. Kipf and M. Welling, “Semi-supervised classification with graph convolutional networks,”arXiv preprint arXiv:1609.02907, 2016

  21. [29]

    Recognizing action units for facial expression analysis,

    Y .-I. Tian, T. Kanade, and J. F. Cohn, “Recognizing action units for facial expression analysis,”IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 23, no. 2, pp. 97–115, 2001

  22. [30]

    Automatic facial expression recognition using features of salient facial patches,

    S. Happy and A. Routray, “Automatic facial expression recognition using features of salient facial patches,”IEEE transactions on Affective Computing, vol. 6, no. 1, pp. 1–12, 2014

  23. [31]

    Facial expression recognition in image sequences using geometric deformation features and support vector machines,

    I. Kotsia and I. Pitas, “Facial expression recognition in image sequences using geometric deformation features and support vector machines,” IEEE transactions on image processing, vol. 16, no. 1, pp. 172–187, 2006

  24. [32]

    Geometry- aware facial expression recognition via attentive graph convolutional networks,

    R. Zhao, T. Liu, Z. Huang, D. P. Lun, and K.-M. Lam, “Geometry- aware facial expression recognition via attentive graph convolutional networks,”IEEE Transactions on Affective Computing, vol. 14, no. 2, pp. 1159–1174, 2021

  25. [33]

    Facial landmark detection by deep multi-task learning,

    Z. Zhang, P. Luo, C. C. Loy, and X. Tang, “Facial landmark detection by deep multi-task learning,” inComputer Vision–ECCV 2014: 13th Eu- ropean Conference, Zurich, Switzerland, September 6-12,Proceedings, Part VI 13. Springer, 2014, pp. 94–108

  26. [34]

    Transformer embedded spectral- based graph network for facial expression recognition,

    X. Jin, X. Song, X. Wu, and W. Yan, “Transformer embedded spectral- based graph network for facial expression recognition,”International Journal of Machine Learning and Cybernetics, vol. 15, no. 6, pp. 2063– 2077, 2024

  27. [35]

    A joint hierarchi- cal cross-attention graph convolutional network for multi-modal facial expression recognition,

    C. Xu, Y . Du, J. Wang, W. Zheng, T. Li, and Z. Yuan, “A joint hierarchi- cal cross-attention graph convolutional network for multi-modal facial expression recognition,”Computational Intelligence, vol. 40, no. 1, p. e12607, 2024

  28. [36]

    A descriptive human visual cognitive strategy using graph neural network for facial expression recognition,

    S. Liu, S. Huang, W. Fu, and J. C.-W. Lin, “A descriptive human visual cognitive strategy using graph neural network for facial expression recognition,”International Journal of Machine Learning and Cybernet- ics, vol. 15, no. 1, pp. 19–35, 2024

  29. [37]

    Attentional visual graph neural network based facial expression recognition method,

    W. Dong, X. Zheng, L. Zhang, and Y . Zhang, “Attentional visual graph neural network based facial expression recognition method,”Signal, Image and Video Processing, vol. 18, no. 12, pp. 8693–8705, 2024

  30. [38]

    Ter-ca-wgnn: trimodel emotion recognition using cumulative attribute-weighted graph neural network,

    H. F. T. Al-Saadawi and R. Das, “Ter-ca-wgnn: trimodel emotion recognition using cumulative attribute-weighted graph neural network,” Applied Sciences, vol. 14, no. 6, p. 2252, 2024

  31. [39]

    Facial action units as a joint dataset training bridge for facial expression recognition,

    S. Mao, X. Li, F. Zhang, X. Peng, and Y . Yang, “Facial action units as a joint dataset training bridge for facial expression recognition,”IEEE Transactions on Multimedia, 2025

  32. [40]

    Design and research of facial expression recognition system based on key point extraction

    Y . Qu and Y . Liu, “Design and research of facial expression recognition system based on key point extraction.”KSII Transactions on Internet & Information Systems, vol. 19, no. 1, 2025

  33. [41]

    Modeling fine-grained relations in dynamic space-time graphs for video-based facial expression recognition,

    C. Huang, F. Jiang, Z. Han, X. Huang, S. Wang, Y . Zhu, Y . Jiang, and B. Hu, “Modeling fine-grained relations in dynamic space-time graphs for video-based facial expression recognition,”IEEE Transactions on Affective Computing, 2025

  34. [42]

    Facial expression recognition using graph-based features and artificial neural networks,

    C. Tanchotsrinon, S. Phimoltares, and S. Maneeroj, “Facial expression recognition using graph-based features and artificial neural networks,” in2011 IEEE International Conference on Imaging Systems and Tech- niques. IEEE, 2011, pp. 331–334

  35. [43]

    Gcf: Graph convolutional networks for facial expression recognition,

    H. Kassab, M. Bahaa, and A. Hamdi, “Gcf: Graph convolutional networks for facial expression recognition,” in2024 Intelligent Methods, Systems, and Applications (IMSA). IEEE, 2024, pp. 166–171

  36. [44]

    Facial expression recognition based on graph neural network,

    X. Xu, Z. Ruan, and L. Yang, “Facial expression recognition based on graph neural network,” in2020 IEEE 5th International Conference on Image, Vision and Computing (ICIVC). IEEE, 2020, pp. 211–214

  37. [45]

    Dual subspace manifold learning based on gcn for intensity-invariant facial expression recognition,

    J. Chen, J. Shi, and R. Xu, “Dual subspace manifold learning based on gcn for intensity-invariant facial expression recognition,”Pattern Recognition, vol. 148, p. 110157, 2024

  38. [46]

    Facial landmark-based emotion recognition via directed graph neural network,

    Q. T. Ngoc, S. Lee, and B. C. Song, “Facial landmark-based emotion recognition via directed graph neural network,”Electronics, vol. 9, no. 5, p. 764, 2020

  39. [47]

    Modeling relational data with graph convolutional networks,

    M. Schlichtkrull, T. N. Kipf, P. Bloem, R. Van Den Berg, I. Titov, and M. Welling, “Modeling relational data with graph convolutional networks,” inThe semantic web: 15th international conference, ESWC 2018, Heraklion, Crete, Greece, June 3–7,proceedings 15. Springer, 2018, pp. 593–607

  40. [48]

    Graph convolutional networks for text classification,

    L. Yao, C. Mao, and Y . Luo, “Graph convolutional networks for text classification,” inProceedings of the AAAI Conference on Artificial Intelligence, vol. 33, no. 01, 2019, pp. 7370–7377

  41. [49]

    Facial expression recognition from near-infrared videos,

    G. Zhao, X. Huang, M. Taini, S. Z. Li, and M. Pietik ¨aInen, “Facial expression recognition from near-infrared videos,”Image and vision computing, vol. 29, no. 9, pp. 607–619, 2011

  42. [50]

    The enterface’05 audio- visual emotion database,

    O. Martin, I. Kotsia, B. Macq, and I. Pitas, “The enterface’05 audio- visual emotion database,” inProceedings of the 22nd IEEE International Conference on Data Engineering Workshops (ICDEW). IEEE, 2006, pp. 8–8

  43. [51]

    Emotiw 2018: Audio- video, student engagement and group-level affect prediction,

    A. Dhall, A. Kaur, R. Goecke, and T. Gedeon, “Emotiw 2018: Audio- video, student engagement and group-level affect prediction,” inPro- ceedings of the 20th ACM International Conference on Multimodal Interaction, 2018, pp. 653–656

  44. [52]

    Geometry guided adversarial facial expression synthesis,

    L. Song, Z. Lu, R. He, Z. Sun, and T. Tan, “Geometry guided adversarial facial expression synthesis,” inProceedings of the 26th ACM interna- tional conference on Multimedia, 2018, pp. 627–635

  45. [53]

    Mer-gcn: Micro- expression recognition based on relation modeling with graph convolu- tional networks,

    L. Lo, H.-X. Xie, H.-H. Shuai, and W.-H. Cheng, “Mer-gcn: Micro- expression recognition based on relation modeling with graph convolu- tional networks,” inProceedings of the IEEE International Conference on Multimedia Information Processing and Retrieval (MIPR). IEEE, 2020, pp. 79–84

  46. [54]

    Facial expression recognition via deep action units graph network based on psychological mechanism,

    Y . Liu, X. Zhang, Y . Lin, and H. Wang, “Facial expression recognition via deep action units graph network based on psychological mechanism,” IEEE Transactions on Cognitive and Developmental Systems, vol. 12, no. 2, pp. 311–322, 2019

  47. [55]

    Geometry- contrastive gan for facial expression transfer,

    F. Qiao, N. Yao, Z. Jiao, Z. Li, H. Chen, and H. Wang, “Geometry- contrastive gan for facial expression transfer,”arXiv preprint arXiv 1802.01822, 2018

  48. [56]

    3d dense geometry-guided facial expression synthesis by adversarial learning,

    R. Bodur, B. Bhattarai, and T.-K. Kim, “3d dense geometry-guided facial expression synthesis by adversarial learning,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Winter, 2021, pp. 2392–2401

  49. [57]

    Emotion recognition with facial landmark heatmaps,

    S. Mo, W. Yang, G. Wang, and Q. Liao, “Emotion recognition with facial landmark heatmaps,” inMultiMedia Modeling: 26th International Con- ference, MMM 2020, Daejeon, South Korea, January 5–8, Proceedings, Part I 26. Springer, 2020, pp. 278–289

  50. [58]

    Facial expression recognition using enhanced deep 3d convolutional neural networks,

    B. Hasani and M. H. Mahoor, “Facial expression recognition using enhanced deep 3d convolutional neural networks,” inProceedings of the IEEE CVPRW, 2017, pp. 30–40

  51. [59]

    Region attention networks for pose and occlusion robust facial expression recognition,

    K. Wang, X. Peng, J. Yang, D. Meng, and Y . Qiao, “Region attention networks for pose and occlusion robust facial expression recognition,” IEEE Transactions on Image Processing, vol. 29, pp. 4057–4069, 2020

  52. [60]

    Comparison between geometry-based and gabor-wavelets-based facial expression recognition using multi-layer perceptron,

    Z. Zhang, M. Lyons, M. Schuster, and S. Akamatsu, “Comparison between geometry-based and gabor-wavelets-based facial expression recognition using multi-layer perceptron,” inProceedings Third IEEE International Conference on Automatic face and gesture recognition. IEEE, 1998, p...

  53. [61]

    Multi-task learning of facial landmarks and expression,

    T. Devries, K. Biswaranjan, and G. W. Taylor, “Multi-task learning of facial landmarks and expression,” in2014 Canadian conference on computer and robot vision. IEEE, 2014, pp. 98–103

  54. [62]

    Multi-task, multi-label and multi-domain learning with residual convolutional networks for emotion recognition,

    G. Pons and D. Masip, “Multi-task, multi-label and multi-domain learning with residual convolutional networks for emotion recognition,” arXiv preprint arXiv:1802.06664, 2018

  55. [63]

    Multimodal learning for facial expression recognition,

    W. Zhang, Y . Zhang, L. Ma, J. Guan, and S. Gong, “Multimodal learning for facial expression recognition,”Pattern Recognition, vol. 48, no. 10, pp. 3191–3202, 2015

  56. [64]

    A visual attention based roi detection method for facial expression recognition,

    W. Sun, H. Zhao, and Z. Jin, “A visual attention based roi detection method for facial expression recognition,”Neurocomputing, vol. 296, pp. 12–22, 2018

  57. [65]

    Patch-gated cnn for occlusion- aware facial expression recognition,

    Y . Li, J. Zeng, S. Shan, and X. Chen, “Patch-gated cnn for occlusion- aware facial expression recognition,” in2018 24th Proceedings of the International Conference on Pattern Recognition (ICPR). IEEE, 2018, pp. 2209–2214

  58. [66]

    Poster: A pyramid cross-fusion transformer network for facial expression recognition,

    C. Zheng, M. Mendieta, and C. Chen, “Poster: A pyramid cross-fusion transformer network for facial expression recognition,” inProceedings of JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. X, AUGUST 2025 13 the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), ...

  59. [67]

    Transformer-based multimodal emotional perception for dynamic facial expression recogni- tion in the wild,

    X. Zhang, M. Li, S. Lin, H. Xu, and G. Xiao, “Transformer-based multimodal emotional perception for dynamic facial expression recogni- tion in the wild,”IEEE Transactions on Circuits and Systems for Video Technology, 2023

  60. [68]

    Tokens-to-token vit: Training vision transformers from scratch on imagenet,

    L. Yuan, Y . Chen, T. Wang, W. Yu, Y . Shi, Z.-H. Jiang, F. E. Tay, J. Feng, and S. Yan, “Tokens-to-token vit: Training vision transformers from scratch on imagenet,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2021, pp. 558– 567

  61. [69]

    Deeper insights into graph convolutional networks for semi-supervised learning,

    Q. Li, Z. Han, and X.-M. Wu, “Deeper insights into graph convolutional networks for semi-supervised learning,” inProceedings of the AAAI Conference on Artificial Intelligence, vol. 32, no. 1, 2018

  62. [70]

    Dual graph convolutional networks for graph- based semi-supervised classification,

    C. Zhuang and Q. Ma, “Dual graph convolutional networks for graph- based semi-supervised classification,” inProceedings of the ACM Web Conference, 2018, pp. 499–508

  63. [71]

    Facial expression recognition using spatial-temporal semantic graph network,

    J. Zhou, X. Zhang, Y . Liu, and X. Lan, “Facial expression recognition using spatial-temporal semantic graph network,” inProceedings of the IEEE International Conference on Image Processing (ICIP). IEEE, 2020, pp. 1961–1965

  64. [72]

    Learning the connectivity: Situational graph convolution network for facial expression recognition,

    J. Zhou, X. Zhang, and Y . Liu, “Learning the connectivity: Situational graph convolution network for facial expression recognition,” inPro- ceedings of the IEEE International Conference on Visual Communica- tions and Image Processing (VCIP). IEEE, 2020, pp. 230–234

  65. [73]

    dlib c++ library,

    D. E. King, “dlib c++ library,” http://dlib.net/, Accessed: ¡Insert Access Date¿

  66. [74]

    Joint fine-tuning in deep neural networks for facial expression recognition,

    H. Jung, S. Lee, J. Yim, S. Park, and J. Kim, “Joint fine-tuning in deep neural networks for facial expression recognition,” inProceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), 2015, pp. 2983–2991

  67. [75]

    Peak-piloted deep network for facial expression recognition,

    X. Zhao, X. Liang, L. Liu, T. Li, Y . Han, N. Vasconcelos, and S. Yan, “Peak-piloted deep network for facial expression recognition,” in Computer Vision–ECCV 2016: 14th European Conference, Amsterdam, The Netherlands, October 11-14, Proceedings, Part II 14. Springer, 2016, pp. 425–442

  68. [76]

    Deep generative- contrastive networks for facial expression recognition,

    Y . Kim, B. Yoo, Y . Kwak, C. Choi, and J. Kim, “Deep generative- contrastive networks for facial expression recognition,”arXiv preprint arXiv:1703.07140, 2017

  69. [77]

    Facenet2expnet: Regularizing a deep face recognition net for expression recognition,

    H. Ding, S. K. Zhou, and R. Chellappa, “Facenet2expnet: Regularizing a deep face recognition net for expression recognition,” in2017 12th IEEE international conference on automatic face & gesture recognition (FG 2017). IEEE, 2017, pp. 118–126

  70. [78]

    Facial expression recognition by de- expression residue learning,

    H. Yang, U. Ciftci, and L. Yin, “Facial expression recognition by de- expression residue learning,” inProceedings of the IEEE conference on CVPR, 2018, pp. 2168–2177

  71. [79]

    Automatic recognition of facial displays of unfelt emotions,

    K. Kulkarni, C. A. Corneanu, I. Ofodile, S. Escalera, X. Baro, S. Hyniewska, J. Allik, and G. Anbarjafari, “Automatic recognition of facial displays of unfelt emotions,”IEEE transactions on affective computing, vol. 12, no. 2, pp. 377–390, 2018

  72. [80]

    Multimodal infor- mation fusion application to human emotion recognition from face and speech,

    M. Mansoorizadeh and N. Moghaddam Charkari, “Multimodal infor- mation fusion application to human emotion recognition from face and speech,”Multimedia Tools and Applications, vol. 49, no. 2, pp. 277–297, 2010

  73. [81]

    Baum-1: A spontaneous audio-visual face database of affective and mental states,

    S. Zhalehpour, O. Onder, Z. Akhtar, and C. E. Erdem, “Baum-1: A spontaneous audio-visual face database of affective and mental states,” IEEE Transactions on Affective Computing, vol. 8, no. 3, pp. 300–313, 2016

  74. [82]

    Learning affective features with a hybrid deep model for audio–visual emotion recognition,

    S. Zhang, S. Zhang, T. Huang, W. Gao, and Q. Tian, “Learning affective features with a hybrid deep model for audio–visual emotion recognition,” IEEE transactions on circuits and systems for video technology, vol. 28, no. 10, pp. 3030–3043, 2017

  75. [83]

    Holonet: towards robust emotion recognition in the wild,

    A. Yao, D. Cai, P. Hu, S. Wang, L. Sha, and Y . Chen, “Holonet: towards robust emotion recognition in the wild,” inProceedings of the 18th ACM international conference on multimodal interaction, 2016, pp. 472–478

  76. [84]

    Facial expression recognition with deeply-supervised attention network,

    Y . Fan, V . O. Li, and J. C. Lam, “Facial expression recognition with deeply-supervised attention network,”IEEE transactions on affective computing, vol. 13, no. 2, pp. 1057–1071, 2020

  77. [85]

    Deep residual learning for image recognition,

    K. He, X. Zhang, S. Ren, and J. Sun, “Deep residual learning for image recognition,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2016, pp. 770–778

  78. [86]

    Efficientnet: Rethinking model scaling for con- volutional neural networks,

    M. Tan and Q. Le, “Efficientnet: Rethinking model scaling for con- volutional neural networks,” inInternational conference on machine learning. PMLR, 2019, pp. 6105–6114

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.